1. Computer Architecture
A Quantitative Approach, Fifth Edition
Chapter 1
Fundamentals of Quantitative
Design and Analysis
Copyright © 2012, Elsevier Inc. All rights reserved.
1

2. 
Performance improvements:

Improvements in semiconductor technology



Enabled by HLL compilers, UNIX
Lead to RISC architectures
Together have enabled:




Feature size, clock speed
Improvements in computer architectures


Introduction
Computer Technology
Lightweight computers
Productivity-based managed/interpreted programming
languages
SaaS, Virtualization, Cloud
Applications evolution:

Speech, sound, images, video, “augmented/extende
d reality”, “big data”
Copyright © 2012, Elsevier Inc. All rights reserved.
2

3. Move to multi-processor
Introduction
Single Processor Performance
RISC
Copyright © 2012, Elsevier Inc. All rights reserved.
3

4. 
Cannot continue to leverage Instruction-Level
parallelism (ILP)


Single processor performance improvement ended in
2003
New models for performance:




Introduction
Current Trends in Architecture
Data-level parallelism (DLP)
Thread-level parallelism (TLP)
Request-level parallelism (RLP)
These require explicit restructuring of the
application
Copyright © 2012, Elsevier Inc. All rights reserved.
4

5. 
Personal Mobile Device (PMD)



Desktop Computing


Emphasis on availability (very costly
downtime!), scalability, throughput (20 million)
Clusters / Warehouse Scale Computers




Emphasis on price-performance (0.35 billion)
Servers


e.g. smart phones, tablet computers (1.8 billion sold 2010)
Emphasis on energy efficiency and real-time
Classes of Computers
Classes of Computers
Used for “Software as a Service (SaaS)”, PaaS, IaaS, etc.
Emphasis on availability ($6M/hour-downtime at Amazon.com!)
and price-performance (power=80% of TCO!)
Sub-class: Supercomputers, emphasis: floating-point
performance and fast internal networks, and big data analytics
Embedded Computers (19 billion in 2010)

Emphasis: price
Copyright © 2012, Elsevier Inc. All rights reserved.
5

6. 
Classes of parallelism in applications:



Data-Level Parallelism (DLP)
Task-Level Parallelism (TLP)
Classes of Computers
Parallelism
Classes of architectural parallelism:




Instruction-Level Parallelism (ILP)
Vector architectures/Graphic Processor Units (GPUs)
Thread-Level Parallelism
Request-Level Parallelism
Copyright © 2012, Elsevier Inc. All rights reserved.
6

7. 
Single instruction stream, single data stream (SISD)

Single instruction stream, multiple data streams (SIMD)




Vector architectures
Multimedia extensions
Graphics processor units
Multiple instruction streams, single data stream (MISD)


Classes of Computers
Flynn’s Taxonomy
No commercial implementation
Multiple instruction streams, multiple data streams
(MIMD)


Tightly-coupled MIMD
Loosely-coupled MIMD
Copyright © 2012, Elsevier Inc. All rights reserved.
7

8. 
“Old” view of computer architecture:


Instruction Set Architecture (ISA) design
i.e. decisions regarding:


registers, memory addressing, addressing modes,
instruction operands, available operations, control flow
instructions, instruction encoding
Defining Computer Architecture
Defining Computer Architecture
“Real” computer architecture:



Specific requirements of the target machine
Design to maximize performance within constraints:
cost, power, and availability
Includes ISA, microarchitecture, hardware
Copyright © 2012, Elsevier Inc. All rights reserved.
8

9. 
Integrated circuit technology



Transistor density: 35%/year
Die size: 10-20%/year
Integration overall: 40-55%/year

DRAM capacity: 25-40%/year (slowing)

Trends in Technology
Trends in Technology
Flash capacity: 50-60%/year


15-20X cheaper/bit than DRAM
Magnetic disk technology: 40%/year


15-25X cheaper/bit then Flash
300-500X cheaper/bit than DRAM
Copyright © 2012, Elsevier Inc. All rights reserved.
9

10. 
Bandwidth or throughput




Total work done in a given time
10,000-25,000X improvement for processors over the
1st milestone
300-1200X improvement for memory and disks over
the 1st milestone
Trends in Technology
Bandwidth and Latency
Latency or response time



Time between start and completion of an event
30-80X improvement for processors over the 1st
milestone
6-8X improvement for memory and disks over the 1st
milestone
Copyright © 2012, Elsevier Inc. All rights reserved.
10

11. Trends in Technology
Bandwidth and Latency
Log-log plot of bandwidth and latency milestones
Copyright © 2012, Elsevier Inc. All rights reserved.
11

12. 
Feature size



Minimum size of transistor or wire in x or y
dimension
10 microns in 1971 to .032 microns in 2011
Transistor performance scales linearly



Trends in Technology
Transistors and Wires
Wire delay does not improve with feature size!
Integration density scales quadratically
Linear performance and quadratic density
growth present a challenge and
opportunity, creating the need for computer
architect!
Copyright © 2012, Elsevier Inc. All rights reserved.
12

13. 
Problem: Get power in, get power out

Thermal Design Power (TDP)



Characterizes sustained power consumption
Used as target for power supply and cooling system
Lower than peak power, higher than average power
consumption

Clock rate can be reduced dynamically to limit
power consumption

Trends in Power and Energy
Power and Energy
Energy per task is often a better measurement
Copyright © 2012, Elsevier Inc. All rights reserved.
13

14. 
Dynamic energy



Dynamic power


Transistor switch from 0 -> 1 or 1 -> 0
½ x Capacitive load x Voltage2
Trends in Power and Energy
Dynamic Energy and Power
½ x Capacitive load x Voltage2 x Frequency switched
Reducing clock rate reduces power, not energy
Copyright © 2012, Elsevier Inc. All rights reserved.
14

15. 



Intel 80386
consumed ~ 2 W
3.3 GHz Intel
Core i7 consumes
130 W
Heat must be
dissipated from
1.5 x 1.5 cm chip
This is the limit of
what can be
cooled by air
Copyright © 2012, Elsevier Inc. All rights reserved.
Trends in Power and Energy
Power
15

16. 
Techniques for reducing power:




Do nothing well
Dynamic Voltage-Frequency Scaling
Low power state for DRAM, disks
Overclocking, turning off cores
Copyright © 2012, Elsevier Inc. All rights reserved.
Trends in Power and Energy
Reducing Power
16

17. 
Static power consumption





Currentstatic x Voltage
Scales with number of transistors
To reduce: power gating
Race-to-halt
Trends in Power and Energy
Static Power
The new primary evaluation for design
innovation


Tasks per joule
Performance per watt
Copyright © 2012, Elsevier Inc. All rights reserved.
17

18. 
Cost driven down by learning curve

Trends in Cost
Trends in Cost
Yield

DRAM: price closely tracks cost

Microprocessors: price depends on
volume

10% less for each doubling of volume
Copyright © 2012, Elsevier Inc. All rights reserved.
18

19. 
Integrated circuit

Trends in Cost
Integrated Circuit Cost
Bose-Einstein formula:
Defects per unit area = 0.016-0.057 defects per square cm (2010)
N = process-complexity factor = 11.5-15.5 (40 nm, 2010)
The manufacturing process dictates the wafer cost, wafer yield and
defects per unit area
The architect’s design affects the die area, which in turn affects the
defects and cost per die




Copyright © 2012, Elsevier Inc. All rights reserved.
19

20. 
Systems alternate between two states of service
with respect to SLA/SLO:
1.
2.

Dependability
Dependability
Service accomplishment, where service is delivered
as specified by SLA
Service interruption, where the delivered service is
different from the SLA
Module reliability: “failure(F)=transition from 1 to
2” and “repair(R)=transition from 2 to 1”




Mean time to failure (MTTF)
Mean time to repair (MTTR)
Mean time between failures (MTBF) = MTTF + MTTR
Availability = MTTF / MTBF
Copyright © 2012, Elsevier Inc. All rights reserved.
20

21. 
Typical performance metrics:



Speedup of X relative to Y


Execution timeY / Execution timeX
Execution time



Response time
Throughput
Measuring Performance
Measuring Performance
Wall clock time: includes all system overheads
CPU time: only computation time
Benchmarks




Kernels (e.g. matrix multiply)
Toy programs (e.g. sorting)
Synthetic benchmarks (e.g. Dhrystone)
Benchmark suites (e.g. SPEC06fp, TPC-C)
Copyright © 2012, Elsevier Inc. All rights reserved.
21

22. 
Take Advantage of Parallelism


e.g. multiple processors, disks, memory
banks, pipelining, multiple functional units
Principle of Locality


Principles
Principles of Computer Design
Reuse of data and instructions
Focus on the Common Case

Amdahl’s Law
Copyright © 2012, Elsevier Inc. All rights reserved.
22

23. 
Principles
Principles of Computer Design
The Processor Performance Equation
Copyright © 2012, Elsevier Inc. All rights reserved.
23

24. 
Principles
Principles of Computer Design
Different instruction types having different
CPIs
Copyright © 2012, Elsevier Inc. All rights reserved.
24

25. Chapter 1 Review & Examples
Copyright © 2012, Elsevier Inc. All rights reserved.
25

26. Instruction Set Architecture (ISA)
• Serves as an interface between software and
hardware.
• Provides a mechanism by which the software
tells the hardware what should be done.
High level language code : C, C++, Java, Fortran,
compiler
Assembly language code: architecture specific statements
assembler
Machine language code: architecture specific bit patterns
software
instruction set
hardware
CSCE430/830
ISA

27. Instruction Set Design Issues
• Instruction set design issues include:
– Where are operands stored?
» registers, memory, stack, accumulator
– How many explicit operands are there?
» 0, 1, 2, or 3
– How is the operand location specified?
» register, immediate, indirect, . . .
– What type & size of operands are supported?
» byte, int, float, double, string, vector. . .
– What operations are supported?
» add, sub, mul, move, compare . . .
CSCE430/830
ISA

28. Classifying ISAs
Accumulator (before 1960, e.g. 68HC11):
1-address
add A
acc acc + mem[A]
Stack (1960s to 1970s):
0-address
add
tos tos + next
Memory-Memory (1970s to 1980s):
2-address
3-address
add A, B
add A, B, C
mem[A] mem[A] + mem[B]
mem[A] mem[B] + mem[C]
Register-Memory (1970s to present, e.g. 80x86):
2-address
add R1, A
load R1, A
R1 R1 + mem[A]
R1 mem[A]
Register-Register (Load/Store, RISC) (1960s to present, e.g.
MIPS):
3-address
CSCE430/830
add R1, R2, R3
load R1, R2
store R1, R2
R1 R2 + R3
R1 mem[R2]
mem[R1] R2
ISA

29. Operand Locations in Four ISA Classes
GPR
CSCE430/830
ISA

30. Code Sequence C = A + B
for Four Instruction Sets
Stack
Accumulator
Push A
Push B
Add
Pop C
Load A
Add B
Store C
memory
CSCE430/830
acc = acc + mem[C]
Register
(register-memory)
Load R1, A
Add R1, B
Store C, R1
memory
R1 = R1 + mem[C]
Register (loadstore)
Load R1,A
Load R2, B
Add R3, R1, R2
Store C, R3
R3 = R1 + R2
ISA

31. Types of Addressing Modes (VAX)
Addressing Mode
1. Register direct
2. Immediate
3. Displacement
4. Register indirect
5. Indexed
6. Direct
7. Memory Indirect
8. Autoincrement
Action
R4 <- R4 + R3
R4 <- R4 + 3
R4 <- R4 + M[100 + R1]
R4 <- R4 + M[R1]
R4 <- R4 + M[R1 + R2]
R4 <- R4 + M[1000]
R4 <- R4 + M[M[R3]]
R4 <- R4 + M[R2]
R2 <- R2 + d
9. Autodecrement
Add R4, (R2)R4 <- R4 + M[R2]
R2 <- R2 - d
10. Scaled
Add R4, 100(R2)[R3] R4 <- R4 +
M[100 + R2 + R3*d]
• Studies by [Clark and Emer] indicate that modes 1-4 account for
93% of all operands on the VAX.
CSCE430/830
Example
Add R4, R3
Add R4, #3
Add R4, 100(R1)
Add R4, (R1)
Add R4, (R1 + R2)
Add R4, (1000)
Add R4, @(R3)
Add R4, (R2)+
ISA

32. Types of Operations
•
•
•
•
•
•
•
•
CSCE430/830
Arithmetic and Logic:
Data Transfer:
Control
System
Floating Point
Decimal
String
Graphics
AND, ADD
MOVE, LOAD, STORE
BRANCH, JUMP, CALL
OS CALL, VM
ADDF, MULF, DIVF
ADDD, CONVERT
MOVE, COMPARE
(DE)COMPRESS
ISA

33. MIPS Instructions
• All instructions exactly 32 bits wide
• Different formats for different purposes
• Similarities in formats ease implementation
6 bits
31
31
5 bits
5 bits
5 bits
5 bits
op
rs
rt
rd
6 bits
5 bits
5 bits
16 bits
op
rs
rt
offset
shamt funct
6 bits
31
CSCE430/830
6 bits
address
0
I-Format
26 bits
op
0
R-Format
J-Format
0
ISA-2

34. MIPS Instruction Types
• Arithmetic & Logical - manipulate data in
registers
add $s1, $s2, $s3
or $s3, $s4, $s5
$s1 = $s2 + $s3
$s3 = $s4 OR $s5
• Data Transfer - move register data to/from
memory  load & store
lw $s1, 100($s2)
sw $s1, 100($s2)
$s1 = Memory[$s2 + 100]
Memory[$s2 + 100] = $s1
• Branch - alter program flow
beq $s1, $s2, 25
if ($s1==$s1) PC = PC + 4 + 4*25
else PC = PC + 4
CSCE430/830
ISA-2

35. MIPS Arithmetic & Logical
Instructions
• Instruction usage (assembly)
add dest, src1, src2
sub dest, src1, src2
and dest, src1, src2
dest=src1 + src2
dest=src1 - src2
dest=src1 AND src2
• Instruction characteristics
– Always 3 operands: destination + 2 sources
– Operand order is fixed
– Operands are always general purpose registers
• Design Principles:
– Design Principle 1: Simplicity favors regularity
– Design Principle 2: Smaller is faster
CSCE430/830
ISA-2

36. Arithmetic & Logical Instructions Binary Representation
6 bits
5 bits
5 bits
op
31
5 bits
rs
rt
rd
5 bits
6 bits
shamt funct
0
• Used for arithmetic, logical, shift instructions
–
–
–
–
–
–
op: Basic operation of the instruction (opcode)
rs: first register source operand
rt: second register source operand
rd: register destination operand
shamt: shift amount (more about this later)
funct: function - specific type of operation
• Also called “R-Format” or “R-Type”
Instructions
CSCE430/830
ISA-2

37. Arithmetic & Logical Instructions Binary Representation Example
• Machine language for
add $8, $17, $18
• See reference card for op, funct values
6 bits
5 bits
5 bits
op
rs
rt
rd
0
31
5 bits
17
18
8
5 bits
6 bits
shamt funct
0
32
000000 10001 10010 01000 00000 100000
CSCE430/830
0
Decimal
Binary
ISA-2

38. MIPS Data Transfer Instructions
• Transfer data between registers and memory
• Instruction format (assembly)
lw $dest, offset($addr)
sw $src, offset($addr)
load word
store word
• Uses:
– Accessing a variable in main memory
– Accessing an array element
CSCE430/830
ISA-2

39. Review: Chapter 1
• Classes of Computers and Classes of
Parallelism
• Technology Trend
• Dependability
• Performance Measurements and Benchmarks
• Principles
CSCE430/830
ISA-2

40. 5 Classes of Computers
• Personal Mobile Devices
– Cost is its primary concern
– Energy, media performance, and responsiveness
• Desktop Computing
– Price-Performance is its primary concern
• Servers
– Availability, Scalability, and Throughput
• Clusters/warehouse-scale computers
– Price-Performance, Energy
• Embedded Computer
– Price
CSCE430/830
ISA-2

41. Classes of Parallelism & Architectures
• Data-Level Parallelism
– Data items can be operated on at the same time
• Task-Level Parallelism
– Tasks can operate independently and largely in parallel
• Instruction-Level Parallelism: data-level para.
– Pipelining, speculative execution
• Vector Architectures & GPU: data-level para.
– A single instruction operates a collection of data in para.
• Thread-Level Parallelism: either data-level
para. or task-level para.
– Exploits parallelism via parallel threads
• Request-Level Parallelism: task-level para.
CSCE430/830
– Exploits parallelism via decoupled tasks
ISA-2

42. 4 ways for hardware to support
parallelism
• Single Instruction stream, Single Data stream
– SISD
• Single Instruction stream, Multiple Data streams
– SIMD, e.g., GPU, targets data-level parallelism
• Multiple Instruction streams, Single Data stream
– MISD, no commercial multiprocessor of this type
• Multiple Instruction streams, Multiple Data streams
– MIMD, e.g., multi-core processors, targets task-level parallelism
CSCE430/830
ISA-2

43. Trend in Technology
• Integrated Circuit (IC) logic technology
– Moore’s Law: a growth rate in transistor count on a chip
of about 40%-55% per year, or doubling every 18 or 24
months.
• Semiconductor DRAM
– In 2011, a growth rate in capacity: 25%-40% per year
• Flash
– A growth rate in capacity: 50%-60% per year
• Magnetic Disk
– Since 2004, it has dropped back to 40% per year.
CSCE430/830
ISA-2

44. Trend in Performance
• Bandwidth vs. Latency
– The improvement on Bandwidth is much significant than
that on Latency.
CSCE430/830
ISA-2

45. Growth in Processor Performance
Move to multi-processor
Hurdle: Power
Wall
Lack: Instructionlevel Parallelism
Parallelism: via
Pipelining
RISC
CSCE430/830
Locality: using
Cache
ISA-2

46. An example of Intel 486 CPU
released in 1992，66MHz, w/ L2 Cache， 4.96.3W
CSCE430/830
http://www.cpu-world.com/CPUs/80486/IntelA80486DX2-66.html
ISA-2

47. A CPU fan for Intel 486 CPU
CSCE430/830
http://www.cnaweb.com/486-ball-bearing-cpufan.aspx
ISA-2

48. An example of Intel Pentium 4 CPU
released in 2002, 2.8GHz, w/ 512KB Cache, 68.4W
CSCE430/830
http://www.pcplanetsystems.com/abc/product_detail
s.php?item_id=146&category_id=61
ISA-2

49. A typical CPU fan for Intel Pentium 4
http://www.dansdata.com/p4coc.htm
CSCE430/830
ISA-2

50. A special CPU fan for gaming/multimedia users
CSCE430/830
http://www.pcper.com/reviews/Cases-andCooling/Asus-Star-Ice-CPU-Cooler-Review
ISA-2

51. Trend in Power and Energy in IC
• Energydynamic
–
½ X Capacitive Load X Voltage2
• Powerdynamic
–
½ X Capacitive Load X Voltage2 X Freq. switched
• Example
– Intel 486 66MHz Voltage: 5V
– Intel Pentium 4 2.8GHz Voltage: 1.5V
– Intel Core 990x 3.4GHz Voltage: 0.8-1.375V
• Improving Energy Efficiency
– Do nothing well; Dynamic Voltage-Frequency
Scaling(DVFS); Design for typical case; Overclocking
• Powerstatic
–
CSCE430/830
Currentstatic X Voltage
ISA-2

52. Dependability
• Service Accomplishment & Service Interruption
• Transitions between 2 states: Failure & Restoration
• Measurements
– Reliability: a measure of the continuous service accomplishment
from a reference initial instant.
» MTTF: Mean time to failure
» FIT: failures per billion hours, 1/MTTF X 109
» MTTR: Mean time to repair
» MTBF: Mean time between failures = MTTF + MTTR
– Availability: a measure of the service accomplishment with
respect to the alternation between the two states.
» MTTF/(MTTF+MTTR)
» Upper bound: 100%
CSCE430/830
ISA-2

53. Performance Measurements and
Benchmarks
• Metrics
– Throughput: a total amount of work done in a given time
– Response time (Execution time): the time between the start and
the completion of an event
• Speedup of X relative to Y
– Execution timeY / Execution timeX
• Execution time
– Wall clock time: a latency to complete a task
– CPU time: only computation time
• Benchmarks
– Kernels, Toy programs, Synthetic benchmarks
– Benchmark suites: SPEC [CPU] & TPC [Transaction Processing]
– SpecRatio = Execution Timereference / Execution Timetarget
CSCE430/830
ISA-2

54. Design Principles
• Take Advantage of Parallelism
• Principle of Locality
• Focus on the Common Case
– Amdahl’s Law
– Upper bound of the speedup: ?
CSCE430/830
ISA-2

55. Example: Laundry Room
Dirty Laundry
Washing Machine
30 minutes washing
CSCE430/830
Drying Machine
Clean Laundry
90 minutes drying
Total Execution Time: 30+90 = 120 minutes
Washing Portion: 30/120 = ¼
Drying Portion: 90/120 = ¾
ISA-2

56. If we can have two drying machines
Dirty Laundry
Washing Machine
30 minutes washing
CSCE430/830
2 Drying Machines Clean Laundry
90/2=45 minutes
drying
ISA-2

57. Speedup: (30+90)/(30+45)=1.6
Dirty Laundry
Washing Machine
30 minutes washing
CSCE430/830
2 Drying Machines Clean Laundry
90/2=45 minutes
drying
ISA-2

58. If we can have unlimited drying
machines
Dirty Laundry
Washing Machine
30 minutes washing
CSCE430/830
∞ Drying Machines Clean Laundry
? minutes drying
ISA-2

59. Speedup: (30+90)/(30+0)=4
Dirty Laundry
Washing Machine
30 minutes washing
CSCE430/830
∞ Drying Machines Clean Laundry
90/∞ ≈ 0 minutes
drying
ISA-2

60. Design Principles
• Take Advantage of Parallelism
• Principle of Locality
• Focus on the Common Case
– Amdahl’s Law
– Upper bound of the speedup:
» 1 / (1 - Fractionenhanced)
CSCE430/830
ISA-2

61. Exercise 1
• If the new processor is 10 times faster than
the original process, and we assume that the
original processor is busy with computation
40% of the time and is waiting for I/O 60% of
the time, what is the overall speedup gained
by incorporating the enhancement?
• Fractionenhanced = 0.4, Speedupenhanced = 10
• Speedupoverall = 1/(0.6+0.4/10) = 1.56
• What is the upper bound of the overall
speedup?
• Upper bound = 1/0.6 = 1.67
CSCE430/830
ISA-2

62. Exercise 2
• In a disk subsystem:
–
–
–
–
–
10 disks, each rated at 1,000,000-hour MTTF
1 ATA controller, 500,000-hour MTTF
1 power supply, 200,000-hour MTTF
1 fan, 200,000-hour MTTF
1 ATA cable, 1,000,000-hour MTTF
• Assuming the lifetimes are exponentially
distributed and that failures are independent,
compute the MTTF of the system as a whole
CSCE430/830
ISA-2

63. Exercise 2
• Because the overall failure rate of the
collection is the sum of the failure rates of the
modules, the failure rate of the system
– = 10*(1/1,000,000) + 1/500,000 + 1/200,000 + 1/200,000 +
1/1,000,000
– = 23/1,000,000 or 23,000 FIT
• Because MTTF is the inverse of the failure
rate
– MTTFsystem = 1/(23/1,000,000) = 43,500 hours
CSCE430/830
ISA-2